Trying graph-based visualization on U2OS data.

Setup and get data

Normalization and dimensional reduction

PCA:

UMAP:

Velocity

Calculating velocity using velocyto and projecting on PCA and UMAP embeddings

Subsample

Downsample to make things faster..

Calculate scores of observed and projected states in PCA space:

FDG Embedding

Construct FDG embedding using PC1 and PC2
Distance metric: L2, similarity: cosine, k=20, similarity threshold=0.25

Compare FDG to PCA and UMAP embeddings

Make PCA and UMAP embedding for sample subsample used for FDG

## delta projections ... sqrt knn ... transition probs ... done
## calculating arrows ... done
## grid estimates ... grid.sd= 0.1168832  min.arrow.size= 0.002337664  max.grid.arrow.length= 0.0610458  done

## delta projections ... sqrt knn ... transition probs ... done
## calculating arrows ... done
## grid estimates ... grid.sd= 0.09383832  min.arrow.size= 0.001876766  max.grid.arrow.length= 0.0610458  done

Calculate consistency score for all embeddings

## [1] "Mean consistency score for PCA, UMAP, FDG embeddings"
## [1] 0.400676
## [1] 0.3969256
## [1] 0.3708876
## [1] "Median consistency score for PCA, UMAP, FDG embeddings"
## [1] 0.4089879
## [1] 0.4014279
## [1] 0.3751183

Varying graph parameters

Calculating consistency scores for FDG embeddings constructed with different graph parameters (k, similarity threshold):
Number of out edges k:

Similarity threshold: